Back office ops · Production

How Vercel Uses Notion AI Agents to Scale Launches 35% Faster

The problem

Vercel's launch database grew to include dozens of properties manageable only via a form-based intake, while agent prompts were stored in GitHub, requiring engineering involvement and full deployments to update, making iteration slow and excluding non-engineers from adjusting business logic.

First attempt

A Notion form with conditional logic reduced friction but remained form-based. Agent prompts buried in GitHub required pull requests, reviews, and full deployments for any behavior change, blocking non-engineers from updating business logic.

Workflow diagram · grounded in source
1
Ship agent creates launch entries
ai_action
“you give it what you have, such as a name, a date, or a link to whatever work is ready, and it figures out the rest by flagging what you told it directly, what it inferred, and what it is still unsure about. It then creates an entry and …”
2
Ship-DX creates Linear issues
integration
“When a new launch is created or updated in the Launch Calendar, ShipDX automatically creates a corresponding Linear issue for Vercel's developer experience (DX) teams, including docs, community, and others. It pulls details from Notion, …”
3
Ship-DX keeps Linear in sync
integration
“When things change in Notion, Ship-DX comes back and updates the Linear tickets to match, plus a comment explaining what changed and why”
4
Ship Closer verifies daily launches
ai_action
“Ship Closer wakes up each morning, checks Vercel's public changelog against the Launch Calendar, and tries to reason whether a given launch actually happened on the day it was supposed to. If confident, it marks the entry as shipped. If …”
5
Prompts edited directly in Notion
feedback_loop
“The team's solution was to move all user-configurable prompts into Notion. Now a sales leader can open a doc, edit the prompt directly, and the agent picks it up on its next run without requiring engineering involvement”
Reported outcome

Vercel achieved 35% faster shipping and teams reclaim up to nine hours weekly per employee, with 89% of employees reporting increased confidence in shipped product quality.
The agent prompt iteration cycle dropped from roughly a business day to five minutes.

Reported metrics
Shipping speed35% faster shipping
Weekly time reclaimed per employeeup to nine hours weekly per employee
Employee confidence in shipped product quality89%
Prompt iteration cycle timedropped from roughly a business day to five minutes
Reported stack
NotionSlackLinearGitHubGraphQLLinear MCPNotion Worker
Source
https://www.notion.so/customers/vercel
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Vercel achieved 35% faster shipping and teams reclaim up to nine hours weekly per employee, with 89% of employees reporting increased confidence in shipped product quality.

What tools did this team use?

Notion, Slack, Linear, GitHub, GraphQL, Linear MCP, Notion Worker.

What results were reported?

Shipping speed: 35% faster shipping; Weekly time reclaimed per employee: up to nine hours weekly per employee; Employee confidence in shipped product quality: 89%; Prompt iteration cycle time: dropped from roughly a business day to five minutes (source-reported, not independently verified).

What failed first in this deployment?

A Notion form with conditional logic reduced friction but remained form-based.

How is this back office ops AI workflow structured?

Ship agent creates launch entries → Ship-DX creates Linear issues → Ship-DX keeps Linear in sync → Ship Closer verifies daily launches → Prompts edited directly in Notion.

Related back office ops cases